Direct answer — How do you transcribe a podcast on a Mac and turn it into show notes? Drop the finished episode file into an on-device transcription app, let your Mac produce a timestamped transcript, correct names and jargon inline, then export the format each destination needs: Markdown for show notes, SRT or VTT for subtitles and platform uploads, plain text for search. Everything runs locally, so an unreleased episode stays unreleased.
Every finished episode already holds its own show notes, pull quotes and search index, locked inside an audio file. Podcast transcription software on a Mac unlocks them.
The question in 2026 is not whether a machine can transcribe your episode — that was settled years ago — but whether the transcript comes back as a file you own, in the formats your publishing chain consumes, and whether you can transcribe a podcast offline rather than posting an unreleased episode to someone else’s server.
This guide covers the routes for turning podcast audio to text on a Mac, why your own transcript still matters when Apple and Spotify generate one, the pass from episode file to show notes, and what to expect from accuracy on multi-speaker interviews. It assumes the recording is finished: this is post-production, not live dictation.
Why keep your own transcript when Apple and Spotify already make one?
Because a platform transcript is a listener feature, not a production file. It is displayed inside one app, on that platform’s terms, and it is not the document you paste into your CMS.
Apple’s announcement of transcripts for Apple Podcasts states that they “will automatically be available for new episodes shortly after episodes are published” — and leaves the door open for producers: “creators who prefer to provide their own transcripts can do so through RSS tags or in Apple Podcasts Connect for subscriber episodes”.
Spotify’s creator documentation is more explicit about the limits. Transcripts “aren’t available to all creators or in the Spotify for Creators mobile app”. An uploaded transcript “needs to be a VTT or SRT file”, with a maximum size of 5MB. And crucially: “you can’t edit transcripts directly in Spotify for Creators” — the documented workflow is to download the VTT file, “make edits to the file on your device”, then re-upload it.
Read together, the conclusion is the same: the platforms expect a corrected transcript produced somewhere else. Local transcription sits upstream of their pipeline rather than duplicating it.
Four reasons to hold the master file yourself:
- Show notes live on your site. A platform transcript does nothing for a page you control.
- One episode, many artefacts. Social quotes, a newsletter excerpt, chapter descriptions and clip selection all come from the same text.
- The RSS tag is yours to fill. The podcast namespace defines a transcript tag that “is used to link to a transcript or closed captions file”, with
urlandtypeattributes, and notes that “multiple tags can be present for multiple transcript formats”. - Privacy before publication. Embargoed guests and uncleared sponsor reads do not belong in a third-party account.
What are the routes from podcast audio to text on a Mac?
Four, and they differ less in the quality of the transcript than in where your audio goes and how much tooling you maintain.
| Route | Where the audio goes | What you get back | Best for |
|---|---|---|---|
| Cloud transcription service | Uploaded to the vendor | Web editor, exports vary by plan | Teams already working inside that tool |
| Platform auto-transcript | Already on the platform | In-app display; Spotify allows a VTT download | Listeners, not producers |
| Command-line engine (Whisper, whisper.cpp) | Stays on your Mac | Text you script into shape yourself | People who enjoy maintaining a pipeline |
| On-device Mac app | Stays on your Mac | Editable transcript plus exports | A repeatable weekly episode workflow |
The scripted route deserves credit: local engines are free, excellent and under your control. They also make you the maintainer of a conversion step, a model download and a loop, every week, for as long as the show runs.
The Mac has a crowded field of on-device transcription apps, MacWhisper, Aiko, Audido and Whisper Notes among them. They differ less in engine than in workflow: long-file behaviour, whether a queue exists, and whether last month’s transcripts are still findable.
How do you turn an episode into show notes, step by step?
Six steps, and only two of them are the transcription. The rest is deciding what each destination needs before you export anything.
Weesper Transcribe is a Mac App Store app for exactly this stage: existing audio and video files in, editable text out, with nothing uploaded.
- Start from the finished cut, not the raw session, so the timestamps match the published timeline. MP3, WAV, M4A, MP4, MOV and any other file macOS can read are accepted.
- Drop the file in. Transcription runs on your Mac, with Metal acceleration on Apple Silicon. No account, no upload, no round trip.
- Set the source language. Transcription covers about 57 languages with auto-detection; fixing the language explicitly beats letting detection guess on a bilingual intro.
- Correct the names inline. Guest names, sponsors, product names and in-jokes are the words no general model has seen. Clickable timestamps and a built-in audio player let you jump to the moment rather than scrubbing for it.
- Export Markdown for the show notes. What you get is a clean, corrected transcript — the raw material for the summary, the chapter list and the quotes. It is not an automatic summary and does not pretend to be one: the judgement about what mattered stays yours.
- Export SRT or VTT for everything that plays. That is what the platforms accept and what your video editor expects.
Two limits worth knowing first. The free tier uses the same engine as Pro but, per the App Store listing, “transcribes files up to 15 minutes, one at a time, to plain text” — enough to test the engine on a real file, not enough for a full episode. A one-time Pro purchase “unlocks unlimited length, batch processing, subtitle and document exports (SRT/VTT/DOCX/PDF), transcript editing, full-text and semantic search, clickable timestamps, and a local automation API for scripted workflows”. The app requires macOS 13 or later, on Apple Silicon or on Intel Macs with AVX2 support.
If the one-time model is itself the reason you are looking, we covered why lifetime pricing is spreading across transcription apps separately.
Which export format does each destination need?
One transcript, several files. Each destination wants a specific format, and exporting the right one removes a conversion step later.
| Destination | Format | Why |
|---|---|---|
| Show notes page on your site | Markdown | Headings and links survive the paste into any CMS |
| Spotify transcript upload | VTT or SRT | Spotify requires a “VTT or SRT file”, max 5MB |
| RSS transcript tag | VTT, SRT, TXT, HTML or JSON | The namespace documents text/vtt, application/x-subrip, text/plain, text/html and application/json |
| Video version of the episode | SRT or VTT | Time-coded, drops straight into the editor |
| Sponsor or client deliverable | DOCX or PDF | A document, not a data file |
| Your own archive or analysis | TXT, CSV or JSON | Machine-readable, easy to grep or parse |
WebVTT is worth defining, because most of this chain settles on it. The W3C specification describes “the Web Video Text Tracks format”, whose “main use is for marking up external text track resources in connection with the HTML <track> element” — captions, subtitles, chapters and time-aligned metadata. SubRip (SRT) is the older, simpler time-coded sibling, carried in the podcast namespace as application/x-subrip.
Weesper Transcribe exports nine formats: TXT, Markdown, SRT, VTT, DOCX, PDF, HTML, CSV and JSON. Pick the export targets once per show, not once per episode.
Why does a searchable archive matter more than any single transcript?
Because the value of a podcast archive compounds, and an unsearchable archive compounds nothing. By episode fifty, the question is never “what did we say in episode fifty” — it is “which episode was that in”.
Two kinds of search answer two different questions:
- Keyword search finds the exact string: a guest’s surname, a sponsor mention, a phrase you know you used.
- Semantic search finds a passage when you remember the idea but not the wording, which is how anyone actually remembers a conversation from eight months ago — see finding a passage by describing it rather than quoting it.
A back catalogue is a one-off job rather than a weekly one. Drop the whole folder of past episodes into a queue, the same approach as queueing a folder of recordings in one batch, and every future search covers the entire show rather than the last few months.
What accuracy should you expect on a podcast interview?
Lower than on a clean solo read, and the gap is wider than most people expect. The recording decides the transcript far more than the app does.
NVIDIA’s model card for parakeet-tdt-0.6b-v3 makes the size of the effect concrete. On English-only evaluation sets, the same model reports:
| Test set (English) | Word error rate | What it contains |
|---|---|---|
| LibriSpeech test-clean | 1.93% | Read speech, clean audio |
| AMI | 11.31% | Multi-speaker meeting recordings |
| Earnings-22 | 11.42% | Earnings calls |
Those are English-only figures, from one model, measured on standard benchmark sets rather than on your episode. They describe the material, not the app: a solo monologue on a good microphone behaves like the first row, a remote two-guest conversation with crosstalk like the second.
Three habits that move accuracy more than switching tools:
- Record each speaker on a separate track and transcribe the tracks separately when attribution matters. Nothing in an automatic transcript reliably labels who spoke.
- Fix the vocabulary once per show, not once per episode. The same recurring names fail the same way every week.
- Budget the editing pass. An interview transcript is a first draft; treating it as finished is what makes transcripts look unreliable.
Frequently asked questions
What is the best way to transcribe a podcast on a Mac?
Hand the finished episode file to an on-device transcription app and keep the transcript as a working file. That avoids an upload, returns a transcript you can edit before anyone reads it, and exports into the formats your publishing chain consumes. A local engine such as Whisper or whisper.cpp reaches the same transcript, but you own the conversion step, the model download and the loop.
Can I transcribe a podcast offline, without uploading the audio?
Yes, provided the engine runs on your machine rather than behind an API. A local model transcribes an episode with no network involved, which matters most before publication: unreleased episodes, embargoed guests and uncleared sponsor reads. Weesper Transcribe runs on your Mac and uploads nothing.
Do I still need my own transcript if Apple Podcasts generates one?
Usually yes, because a platform transcript is a listener feature rather than a production file. Apple states that creators “who prefer to provide their own transcripts can do so through RSS tags or in Apple Podcasts Connect for subscriber episodes”, and Spotify says outright that “you can’t edit transcripts directly in Spotify for Creators”.
What format should a podcast transcript be in for show notes?
Markdown for the show notes, and a time-coded subtitle file for anything that plays. The podcast namespace documents a transcript tag carrying mime types including text/plain, text/html, text/vtt, application/json and application/x-subrip, while Spotify specifies that an uploaded transcript “needs to be a VTT or SRT file” with a maximum size of 5MB.
How accurate is automatic transcription on a multi-speaker podcast interview?
Noticeably less accurate than on a clean solo read. NVIDIA’s model card for parakeet-tdt-0.6b-v3 reports, on English evaluation sets, a word error rate of 1.93% on LibriSpeech test-clean — read speech in clean conditions — against 11.31% on AMI, which is multi-speaker meeting audio. That swing is driven by the recording, not the software.
How do I make a back catalogue of episodes searchable?
Transcribe the catalogue once as a batch, then search the transcripts rather than the audio. Queueing a folder turns a hundred separate jobs into one unattended run, and the result is a text archive you can query by keyword or by meaning.
Conclusion
Podcast transcription stopped being a transcription problem some time ago. It is a workflow problem: where the audio goes, who owns the corrected file, and which formats come out the other end.
The platforms generate something useful for listeners, but it is no substitute for a master transcript you can edit, repurpose and publish on your own pages — which is exactly what Apple and Spotify assume you will bring them. Keep the transcription local, correct it once, export into the formats each destination expects, and let the archive accumulate.
Got an episode waiting to be transcribed? Read the full feature list and free-tier limits, or get the app from the Mac App Store — free to download, with a one-time Pro upgrade and no subscription. Setup questions are answered in the support documentation.